When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth Labels

Sushant Gautam, Finn Schwall, Annika Willoch Olstad, Fernando Vallecillos Ruiz, Birk Torpmann-Hagen, Sunniva Maria Stordal Bjørklund, Leon Moonen, Klas Pettersen, Michael A. Riegler

arXiv:2605.06652 · 2026-05-06 공개 · arXiv · PDF

Abstract

Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this setting as benchmarkless comparative safety scoring and specify the contract under which a scenario-based audit can be interpreted as deployment evidence. Scores are valid only under a fixed scenario pack, rubric, auditor, judge, sampling configuration, and rerun budget. Because no labels are available, we replace ground-truth agreement with an instrumental-validity chain: responsiveness to a controlled safe-versus-abliterated contrast, dominance of target-driven variance over auditor and judge artifacts, and stability across reruns. We instantiate the chain in SimpleAudit, a local-first scoring instrument, and validate it on a Norwegian safety pack. Safe and abliterated targets separate with AUROC values between 0.89 and 1.00, target identity is the dominant variance component ($η^2 \approx 0.52$), and severity profiles stabilize by ten reruns. Applying the same chain to Petri shows that it admits both tools. The substantial differences arise upstream of the chain, in claim-contract enforcement and deployment fit. A Norwegian public-sector procurement case comparing Borealis and Gemma 3 demonstrates the resulting evidence in practice: the safer model depends on scenario category and risk measure. Consequently, scores, matched deltas, critical rates, uncertainty, and the auditor and judge used must be reported together rather than collapsed into a single ranking.

한국어 요약

📋 한 줄 요약

**[LLM 안전성 평가]** 라벨이 없는 환경에서 LLM 안전성을 비교 평가하기 위한 도구 타당성 체인을 정식화하고 SimpleAudit으로 노르웨이어 안전성 비교에 적용했다.

🎯 핵심 기여도

💡 핵심 아이디어

새 언어·분야·규제 환경에는 라벨된 벤치마크가 없는 경우가 많다. 진실 라벨 일치 대신, 통제된 안전 vs abliterated 대조 반응성·표적 변동의 지배성·재실행 안정성으로 점수의 도구적 타당성을 입증한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 라벨 부재 환경에서도 신뢰할 수 있는 안전성 비교 평가 절차를 제공한다. **한계**: 도구 타당성 체인은 사전 정의된 시나리오 팩과 심사 구성에 의존하며 단일 글로벌 순위로 환원되지 않는다.

🚀 실용적 활용